Online SVM-based personalizing method for the drowsiness detection of drivers.

Inter-driver variation is one of major problems of the drowsiness detecting system-based on physiological signals. This paper proposes an online support vector machine (OSVM)-based method to solve the problem by the inter-driver variation. The method personalizes the drowsiness detecting system for a certain real user using feedback data from the user. The OSVM selects important data in previous training data and retrains itself with new feedback data for the personalization. Two OSVMs having different initial training data are personalized by the feedback data, and a switching method of the two OSVMs is used in the proposed method for low initial error and fast adaptation. Simulation was conducted using the data obtained by a wearable device and an indoor driving simulator, and the usefulness of the proposed method was validated. The detecting accuracy was increased from 72.05 % to 95.66 % on average for 28 subjects. By feedback data and the proposed method, more accurate drowsiness detection will be possible and it will increase the safety of drivers.